ansys cfd validation for civil transport aircraft in high- lift ...cleanup and part-management is...

1

Tenth International Conference on Computational Fluid Dynamics (ICCFD10), Barcelona, Spain, July 9-13, 2018

ICCFD10-071

ANSYS CFD Validation for Civil Transport Aircraft in High-

Lift Configuration Part-1

Krishna Zore1

ANSYS Software Pvt Ltd, Plot no. 34/1, Hinjewadi, Pune - 411057, Maharashtra, India

Shoaib Shah2, John Stokes3

ANSYS Canada Ltd., 1000 Sherbrooke Street West, Suite 2120, Montreal QC, H3A 3G4 Canada

Balasubramanyam Sasanapuri4

ANSYS Software Pvt Ltd, Unit no. 906, The Platina, Gachibowli, Hyderabad 500032, India

Patrick Sharkey5

ANSYS UK Ltd., 97 Jubilee Avenue, Milton Park Oxfordshire OX14 4RW United Kingdom

Abstract: The paper presents a study of aerodynamic flow computations over a civil transport aircraft high-lift configuration model performed using ANSYS Fluent – a cell-centered, Reynolds-averaged Navier-Stokes CFD solver. The computational results obtained on the JAXA Standard Model (JSM)1 are validated against experimental data from JAXA’s Low Speed Wind Tunnel (JAXA-LWT1). In Part 1, the model is studied in a nominal landing configuration with high-lift devices deployed including the support brackets. The study including an under-wing mounted engine-nacelle-pylon on the JSM will be presented in a second paper as Part-2. The JSM CAD geometry cleanup and part-management is performed using ANSYS SpaceClaim Direct Modeler (SCDM) R18.2, and the unstructured hybrid mesh is generated using ANSYS Fluent Meshing R18.2. Comparison between ANSYS Fluent R18.2 computational results and experimental data over a sweep of angles of attack will be shown based on aerodynamic forces, moments and pressure coefficients as well as surface flow visualizations. The fully turbulent results based on the Shear Stress Transport k-ω (SST)5,6 turbulence model and the transition results using SST-Transition (γ-Reθ)7 model will be discussed. Keywords: High-Lift Aerodynamics, Computational Fluid Dynamics (CFD), Computer-Aided Design (CAD), Turbulence Modeling, Transition, American Institute of Aeronautics and Astronautics, Japan Aerospace Exploration Agency (JAXA).

1 Introduction

Development of efficient high-lift devices for take-off and landing are a very important part of the

aircraft design process and have a strong influence on aircraft performance and operational costs.

For a large commercial transport aircraft with twin jet engines, an improvement of 1% in lift-to-

drag ratio at takeoff can provide a 1270 kg (2800-lb) increase in payload, whereas, an improvement

of 1.5% in maximum lift coefficient at landing can provide a 3000 kg (6600-lb) increase in payload1.

Today’s CFD codes are highly reliable and consistent for simplified cruise-type configurations,

1 Application Software Development

2 Senior Software Development

3 Director Software Development

4 Lead Technology Specialist, Senior AIAA Member

5 Lead Software Development

2

however, the complexities associated with the high-lift configurations make it very challenging to

predict high-lift aerodynamic forces accurately2,12,13,15. One of the main challenges in CFD

modelling of high-lift systems is mesh generation, due to the complex multi-element wing system

comprising slat, main-wing and flap elements (in addition to support brackets and numerous small

gaps). Another challenge is the complexity of flow physics itself, with numerous phenomena such

as boundary-layer transition, flow separation, reattachment and wake-boundary layer interaction

all playing an important role2,3,10,11. Careful resolution of such phenomena and together with the

CFD flow solver accuracy are critical to the accurate prediction of aerodynamic forces required for

the development of an efficient aerodynamic design system.

ANSYS Fluent solver’s high-lift aerodynamic prediction capabilities are validated on the JAXA

Standard Model (JSM) in its nominal landing configuration, with high-lift devices including support

structures such as the slat tracks and Flap Track Fairings (FTF) deployed. This model was tested at

the 6.5m x 5.5m JAXA- Low Speed Wind Tunnel (LWT1)9. The geometry, flow operating conditions

and experimental data are taken from NASA’s 3rd AIAA CFD High Lift Prediction Workshop4.

Geometric cleanup, part management and creation of volume mesh refinement regions known as

body of influences (BOIs), are created in ANSYS Spaceclaim Direct Modeler (SCDM) R18.2.

Unstructured surface (triangular facets) and hybrid volume (prisms and tetrahedrons) meshes are

generated using ANSYS Fluent Meshing R18.2. Finally, computational results obtained with

ANSYS Fluent R18.2 using turbulence models both with and without laminar-turbulent transition

are validated against experimental data.

2 Model Geometry Preparation

The JSM in nominal landing configuration (slat and flap deployed at 300) with support brackets

(such as the slat tracks and FTFs) is shown in Figure 1. The geometric reference parameters for the JSM

are outlined in Table 1.

Figure 1: JSM, wing-body configuration

3

Table 1. JSM Geometry parameters.

Geometric parameters Units Mean aerodynamic chord (MAC) 529.2 mm model scale

Wing semi-span 2300.0 mm

Reference area of the semi-span model (Sref/2) 1123300.0 mm2

Moment reference center (MRC) x = 2375.7 mm, y = 0.0 mm, z = 0.0 mm The ‘watertight’ geometry required for CFD mesh generation is created using the multi-purpose

ANSYS 3D SCDM. A built-in repair toolkit with simple and easy-to-use utilities, such as stitching,

and gap and missing faces detection, quickly addresses any minor CAD incongruities and improve

CAD fidelity. The “Design Sketch” tool is used to create Bodies of Influence (BOIs) for subsequent

use during mesh generation, to obtain better volume mesh refinement control. These BOIs are

especially important where the propensity of complex flow behavior is expected around the slat

tracks, FTFs, multi-element wing gaps, wing-body junction, wing-slat-flap tips and in the high-lift

system wakes. Figure 2 shows these refinement regions. The box-shaped farfield domain enclosing the

JSM is created using the “Prepare-Enclosure” utility tool. To avoid farfield boundary effects on the

JSM, the farfield extends to 100 times the MAC length. Additionally, surface grouping is done using

“Groups” utility, so that during mesh generation, local surface mesh size controls can be directly applied

in areas such as the leading and trailing edges of multi-element wing, high curvature areas, wing-slat-

flap tips, wing-body junction, slat tracks, FTFs, fuselage nose and rear. Once the geometry is prepared,

it is saved in SCDOC format (SCDM native file format) and imported into ANSYS Fluent Meshing.

Figure 2: JSM, farfield domain, and BOI (refinement) regions.

4

3 Mesh Generation ANSYS Fluent Meshing is a robust, unstructured hybrid mesh generation tool which can create

meshes of virtually any size or complexity, consisting of tetrahedral, hexahedral, polyhedral,

prismatic, or pyramidal cells.

ANSYS Fluent Meshing can generate unstructured tetrahedral, hex-core, or hybrid volume CFD

meshes, primarily using two approaches: using an existing surface mesh topology from any other

mesh generation tool or using a combination of the “CAD Faceting” and “CFD Surface Mesh”

options to create an initial, ‘watertight’ surface grid. Either approach serves as the basis from which

a high-quality volume mesh can be generated quickly and efficiently.

In this case, the second approach is followed: the JSM geometry is first imported into ANSYS Fluent

Meshing using the “CAD Faceting” option as a “Mesh Object” feature, which results in a rough,

faceted surface representation of the model. While this model is not suitable for volume mesh

generation, it allows for a quick setup of meshing parameters. Local “Scoped Size Functions” are

then defined on the imported model surfaces to provide better control on surface mesh distribution.

Multiple surface mesh size control functions are available such as ‘soft’, ‘hard’, ‘curvature’ and

‘proximity’ based, all together controlling the maximum and minimum surface mesh size, surface

curvature, and the number of cells per gap or between proximity edges/surfaces. Further

adjustments in sizing can be made by computing a “size field” which provides a visual cue for the

surface mesh sizes. On obtaining the desired surface mesh sizes, the “size field” file is saved and will

be used to generate the triangulated surface mesh. This is done by re-importing the JSM CAD with

the “CFD Surface Mesh” option and the “size field” file specified. On completion, the triangular

surface mesh will appear under “Mesh Objects” ready for the volume mesh generation step.

For this study, a hybrid volume mesh consisting of prism layers above the wall boundaries (to

capture the boundary layer) and isotropic tetrahedral cells filling the remaining volume is created.

The prism layers are generated using a “Uniform” growth option by maintaining a constant initial

height of 0.002 mm and a specified prism growth ratio ranging from 1.1 (wing) to 1.2 (body). To

avoid very thin prism elements, the aspect ratio is restricted using a prism control parameter.

Additionally, a “Gap Factor” value is adjusted to maintain enough space for tetrahedral elements in

proximity regions. To save time, prism-layer generation is done on parallel processors using

automatic partitioning. The previously-generated “size field” file can be applied to the tetrahedral

volume mesh sizing. Additional BOI sizing is required for extra refinement in key complex flow

regions identified earlier. The final overall quality of the volume mesh is then improved using the

“Auto Node Move” smoothing operation. The unstructured hybrid surface and volume mesh

resolution are shown in Figure 3 while the mesh metrics for the JSM are shown in Table 2.

Table 2. JSM mesh metrics.

JSM configuration

Number of nodes

Number of cells

Initial wall spacing, ∆y

(normal dist.)

Number of cells on trailing edges

wing-body 125,670,996 289,270,572 0.002 mm 12

6

Figure 3: JSM, Computational mesh - cross-sectional and surface resolutions.

4 Flow Conditions and Solver The wind tunnel flow conditions are outlined in Table 3 including the flow Mach number and

angles of attack (α) sweep. For CFD, all simulations are “free air”, and no wind tunnel walls or

model support systems are included.

Steady-state RANS simulations were performed with ANSYS Fluent R18.2, a cell-centered finite

volume method solver. A pressure-based fully coupled algorithm is employed with second order

upwind and central discretization methods used for convective and diffusion terms, respectively.

The resulting discrete linear system is solved using a point implicit (Gauss-Seidel) linear equation

solver in conjunction with an algebraic multigrid (AMG) method. Turbulence was modelled using

the k-ω Shear Stress Transport (SST)5,6 model, while transition was modeled using the two-equation

SST-Transition model (γ-Reθ)7. Additionally, the models’ constant a1 was modified from 0.31

(default) to 1, as early tests on this case with the current meshes using a1=1 resulted in separation

being delayed to be more consistent with available experimental observations. It should be noted,

however, that using the default SST model required further investigation (not done for this paper)

looking at the effect of mesh resolution21.

7

Table 3. JSM flow parameters.

Mach Number 0.172

Angles of attack (α) 4.36o, 10.47o, 14.54o, 18.58o, 20.59o, and 21.570o

Reynolds Number based on MAC 1.93 million

Reference Static Temperature 551.79oR (33.40oC or 92.12oF)

Reference Static Pressure 747.70 mmHg (14.458 PSI)

5 Results and Discussions

The JSM is solved at six angles of attack (α). The complete ANSYS Fluent solution analysis matrix for

this study is shown in Table 4.

Table 4. JSM solution matrix.

JSM configuration Turbulence models Angles of attacks

Wing-body SST a1=1

4.36o, 10.47o, 14.54o, 18.58o, 20.59o, 21.57o Transition-SST a1=1

The computational results obtained on the JSM are validated against experimental data from JAXA’s

LWT14,9,10.

5.1 Aerodynamic Coefficient Comparisons

Figure 4 shows the CL-α, CD-α, CL-CD and CL-CM plot comparisons for JSM wing-body with

experimental data. Both the computational results with SST a1=1 and SST-Transition a1=1 show good

agreement with the experimental data at α = 18.580 and below.

Both SST a1=1 and SST-Transition a1=1 models predict stall occurring at α = 18.580 with CLmax =

2.746 and CLmax = 2.811, respectively, while the experiment shows stall at α = 20.590 and CLmax = 2.768.

In general, for angles below 18.580, both models predict CL within ~0.1 – 1.0% of the experiment. The

Transition-SST a1=1 model provides an overall better agreement of CL at the non-linear portion of the

CL-α plot, but the overall CL comparisons are generally very close.

The differences in drag between the two models are also generally very close. However, both CD-α, and

CL-CD plots indicate much more pronounced differences between wind tunnel and computational

results, especially at α = 10.470 and higher. In general, the Transition-SST a1=1 model provides slightly

higher drag values compared to SST a1=1. Prior to stall, total drag differs from the experiments by

~10-60 drag counts with both models.

The CL-CM plots show that the nose down pitching moments are lower for the SST al=1 model than for

the Transition-SST a1=1 model, with the latter generally producing better comparisons to experiment.

These observed differences are possibly due to insufficient grid resolution in areas of highly complex

flow structures found in high-lift configurations. These areas include trailing edge wakes, support

bracket wakes, wing-body junction flow, slat and flap cove vortices, wing tip vortices and other

secondary spiral vortices. Some of the key areas where further refinement is needed will be discussed

at the end of this section. Another possible reason for the differences could be the inherent limitation

of linear eddy-viscosity turbulence models, which neglect the non-linear effects of turbulent, secondary

and swirling flows. There is also an important possible impact of setting a1=1 on the eddy viscosity

term: while it prevents the onset of early flow separation (as compared to SST with default a1), it

appears to lead to the higher drag predictions21. These possibilities all point to the need for further

investigation to fully understand the mesh requirements, turbulence modeling needs, and complex flow

8

behavior associated with such multi-element high-lift devices, especially in the presence of supporting

brackets.

(a) CL vs α.

(b) CD vs α.

(c) CL vs CD.

(d) CL vs CM.

Figure 4: JSM, Aerodynamic force coefficients for angles of attack sweep.

5.2 Experimental Oil Flow and Cp Comparisons

Comparisons between the surface flow patterns and Cp plots of the experimental and computational

results for both models provide further insights. The spanwise Cp extraction locations are shown in

Figure 5. At α=4.360, computational results (Figure 8), generally show good agreement with the

experimental Cp data. The SST-Transition a1=1 results generally show improved Cp comparison

close to the suction peaks of all leading edges. The experimental data shows slightly larger suction

peaks on the flap’s leading edges when compared to the computational results, especially on stations

inboard of the Yehudi break, with SST-Transition a1=1 showing better agreement. The better

agreement suggests that flow on the leading edge of the flap transitions from laminar to turbulent,

as the china-clay visualization in (Figure 6) also indicates. The effect of the FTFs on the flaps are also

visible on the china-clay visualization, with the leading edge experiencing turbulent flow, while a spiral

vortex forms at its inboard trailing edge, as seen in the oil-flow visualization of Figure 7. Both these

phenomena are well captured by the SST a1=1 and SST-Transition a1=1 computations, as seen in

Figure 7. Both SST a1=1 and SST-Transition a1=1 show Side of Body (SOB) flow separation similar

9

to the oil flow visualization, seen as a trailing edge vortex on the flap’s upper surface.

Figure 5. JSM Wing Cp Extraction Stations4.

Additionally, the effect of slat support wakes can be seen on the wing surface in both computational

and experimental results. The surface flow patterns in Figure 7 indicate that the flow behavior at the

wing tip matches well with the oil flow visualization. The wing tip vortices are depicted well, forcing

flow coming from upstream and the slat tip to move inward, before straightening out and eventually

getting directed outward to the wing tip’s trailing edge. In addition, the wing tip vortices together with

the outermost outboard slat support wake result in the birth of a small wing trailing edge vortex. The

transition behavior on the slat and flap seen on the china clay visualization (Figure 6) are matched well

with SST-Transition a1=1 model, though the SST-Transition a1=1 model shows a large laminar section

on the leading edge of the mid-wing section, which is absent in the experiment.

10

Figure 6. JSM with nacelle pylon - Visualization of boundary layer transition by china-clay at α=4.360

& comparison with SST-Transition a1=1 computation results (JSM without nacelle pylon).

(a) Exp. Oil Flow.

(b) SST a1=1

(c) Transition SST a1=1.

Figure 7. JSM, Experimental Oil Flow vs. CFD Surface Flow Pattern at α = 4.360.

(a) A-A (eta=0.16).

(b) B-B (eta=0.25).

(c) C-C (eta=0.33).

11

(d) D-D (eta=0.41).

(e) E-E (eta=0.56).

(f) G-G (eta=0.77).

(g) H-H (eta=0.89).

Figure 8. JSM, Cp Plots Comparison with Experimental Measurements at α = 4.360.

Figure 9. JSM with nacelle pylon - Visualization of boundary layer transition by china-clay at

α=10.470 & comparison with SST-Transition a1=1 computation results (JSM without nacelle

pylon).

At α=10.470, both models show similar results to those at α=4.360, with generally larger suction peaks

observed in the former (Figure 11). Additionally, both the experimental oil flow visualization and the

computational surface flow patterns show larger wing tip vortex structures that locally influence the

outboard wing flow patterns resulting in an additional small wing trailing edge spiral vortex structure

forming beside the one seen at α=4.360. The transition behavior on the slat and flap seen on the china

clay visualization (Figure 9) are matched well with SST-Transition a1=1 model except at certain mid-

slat locations where the onset of trailing edge turbulence is observed in the experiment but not in the

12

computation. Also, the china clay visualizations show turbulent inboard wing leading edges, whereas

the SST-Transition simulation indicates large regions of laminar flow – however the simulation does

not include the nacelle, so this may be a cause for this difference.

(a) Exp. Oil Flow.

(b) SST a1=1.

(c) Transition-SST a1=1.


(a) A-A (eta=0.16).

(b) B-B (eta=0.25).

(c) C-C (eta=0.33).

(d) D-D (eta=0.41).

(e) E-E (eta=0.56).

(f) G-G (eta=0.77).

(g) H-H (eta=0.89).


13

At α=14.540, both models show good Cp agreement with experiment (Figure 13) for all stations except

H-H, with the SST-Transition a1=1 showing the better result. Computational surface flow patterns

show flow deceleration near the outboard slat (resulting in smaller suction peaks) due to separation seen

on most of the outboard wing upper surface. There are no experimental pictures available at this α but

based on the α=18.580 experimental oil flow visualizations, this separation should not be observed.

(b) SST a1=1

(c) Transition-SST a1=1


(a) A-A (eta=0.16).

(b) B-B (eta=0.25).

(c) C-C (eta=0.33).

(d) D-D (eta=0.41).

(e) E-E (eta=0.56).

(f) G-G (eta=0.77).

(g) H-H (eta=0.89).


14

In both SST a1=1 and SST-Transition a1=1, the smaller outboard wing trailing edge spiral vortex

formed earlier at α=4.360, appears to have expanded to occupy a majority of the upper outboard

wing surface, and together with the turbulent wakes from the outer most slat support, led to the

outboard wing leading edge separation. This would also explain the higher drag predictions noted

previously in the CD-α, and CL-CD plots. Insufficient mesh resolution seems the most likely cause

for this large unexpected separation to occur. In general, the slat leading edges are now producing

less suction peaks (stations D-D, E-E, G-G and H-H) as compared to experimental Cp plots, with the

SST-Transition a1=1 now depicting transition to turbulence due to laminar separation, a

phenomenon not seen at lower angles of attack. These differences at α=14.540 suggest that further

mesh refinement may be required in multiple areas: at the leading and trailing edges of outboard

multi-element devices, in and around the outermost slat support and its wake, on the slat and wing

tip, and finally around the outboard slat cove. Additional investigations are required to confirm this

premise and improve understanding of the highly complex flow interactions occurring at these

locations.

The computed surface flow patterns for both models at α=18.580 are quite similar to the results at

α=14.540, especially in the outboard section of the wing. Again, good agreement in the pressure

distribution are generally observed for both SST a1=1 and SST-Transition a1=1, except at station H-

H. The experimental oil flow visualization flow structures at the outboard wing separation area

differ from the computational results. A small spiral vortex seen initially as a trailing edge vortex at

α=4.360, has moved and grown in strength. This vortex appears to interact with both the outermost

slat support turbulent wake and the outboard wing trailing edge separation (caused by the larger

wing tip vortices with larger axial spread). Possible reasons for some of these differences are as

discussed previously. At the wing root and close to the wing-body and flap-body junctions,

computational results generally match well, as seen by both the oil flow visualizations and the Cp

plots. Encouragingly, computational results at α=18.580 show the experimentally observed onset of

wing SOB separation and wash out of flap SOB separation (seen earlier at lower angles of attack) at

their respective trailing edges.

Figure 14. JSM with nacelle pylon - Visualization of boundary layer transition by china-clay at

α=18.580 & comparison with SST-Transition a1=1 computation results (JSM without nacelle pylon).

15

Furthermore, the china clay visualizations (Figure 14) show that the SST- Transition a1=1 results agree

well with the laminar and turbulent regions on the leading edges of the multi-element wing, except at

the leading edge of the inboard wing where turbulent flow dominates, contrary to what is observed in

the computational results – again with the presence of the nacelle in the experiments being a possible

factor.

(a) Exp. Oil Flow.

(b) SST a1=1.

(c) Transition-SST a1=1


(a) A-A (eta=0.16).

(b) B-B (eta=0.25).

(c) C-C (eta=0.33).

(d) D-D (eta=0.41).

(e) E-E (eta=0.56).

(f) G-G (eta=0.77).

(g) H-H (eta=0.89).


16

At α=20.590, good agreement in the pressure distribution are generally observed for both SST a1=1

and SST-Transition a1=1, except at station G-G and H-H. Both models capture a similar outboard wing

leading edge separation in addition to a second large wing leading edge separation close to station G-

G, the latter not likely to occur based on α=21.570 experimental oil flow visualizations. This also

explains the lower suction peaks at the slat leading edges (stations G-G and H-H) as compared to

experimental Cp plots.

(b) SST a1=1.



(a) A-A (eta=0.16).

(b) B-B (eta=0.25).

(c) C-C (eta=0.33).

(d) D-D (eta=0.41).

(e) E-E (eta=0.56).

(f) G-G (eta=0.77).

(g) H-H (eta=0.89).


17

Differences noted earlier were primarily believed to be due to insufficient mesh resolution, and that

also holds here. At the wing root and close to the wing-body and flap-body junctions, computational

results generally match well, as seen by both the oil flow visualizations and the Cp plots. The

computational results show increase in spread of the wing SOB separation. The close agreement

between predicted surface pressure distributions at stations A-A and B-B and experimental data are

consistent with this observation.

At α=21.570, the highest angle of attack simulated, experimental oil flow visualizations show that

flow over the wing-body junction and a majority of the wing outboard section (outward of H-H

station) are now separated, resulting in wing stall. The experiment exhibits a wing SOB separation

that has a wide axial spread (starting at the mid-span root wing-body junction and ending further

downstream at the wing’s trailing edge location (upstream of the inboard FTF)) due to the large

spanwise flow movement experienced. The flow separation behavior and vortex structures,

locations and strengths differ between SST a1=1 and SST-Transition a1=1. Some of these flow

phenomena have been captured well by computation and, depending on the turbulence model,

may be closer to experiment. For instance, the SST a1=1 model does a better approximation of the

spread of the wing SOB separation (Figure 19) than the SST-Transition a1=1 model, as also reflected

in the Cp plot comparisons with experimental data of Figure 20 (stations A-A and B-B). However, it

does not capture the same spanwise spread of the wing SOB separation that the experiment shows. On

the other hand, the SST-Transition a1=1 shows better comparisons with experimental oil-flow

visualizations and surface pressure distributions on the mid-wing’s upper surface. While both models

capture similar outboard wing leading edge separation, it does not capture the trailing edge flow

separation or the spiral vortex past the slat support in a similar manner. In addition, both SST a1=1 and

SST-Transition a1=1, predict a second large wing leading edge separation close to station E-E and G-

G respectively, both not seen in experiment. Previously stated conclusions as to the possible cause for

failure to capture these complex flow phenomena, apply here. Central to these conclusions are the need

for further mesh refinement in key areas in order to achieve grid-independent solutions.

(a) Exp. Oil Flow.

(b) SST a1=1.



18

(a) A-A (eta=0.16).

(b) B-B (eta=0.25).

(c) C-C (eta=0.33).

(d) D-D (eta=0.41).

(e) E-E (eta=0.56).

(f) G-G (eta=0.77).

(g) H-H (eta=0.89).


5.3 Mesh Considerations

Accurately predicting complex flow physics is highly dependent on mesh resolution and this is

examined by comparing the solutions on two different meshes at α = 4.360 with SST a1=1 turbulence

model. One solution was obtained using ANSA-supplied mesh21 (results presented at AIAA

AVIATION 2018) and the other from using Fluent Meshing (details presented in this paper).

Viscous wake interaction is one of the most important phenomena for the high-lift configurations.

For example, adverse pressure gradient separation wakes on slat and wing upper surfaces interact

with SOB separation wakes from the support brackets passing through gaps between slat-wing and

wing-flaps over the upper surfaces of wing and flaps. In general, poor resolution of the wakes leads

to their spread being overpredicted, and ultimately lowering the lift and increasing drag. This is

consistent with the results seen in Figure 21 compared to Figure 22, and the higher drag predicted

for the Fluent Mesh (the same observation can also be made further outboard, although this is not

shown here).

The complex recirculation pattern seen in the cove behind the slat is also evident in these images,

and the different meshes clearly have an impact on the predictions here, too, which in turn can

strongly affect the flow impinging on the leading edge of the main wing and then entering the gap

between slat and main wing. The higher resolution of the slat cove regions in Fluent mesh

19

compared to the ANSA mesh is essential to capture the free shear layers and the slat core vortex in the

slat cavity. The higher eddy viscosity in slat cove region is indicative of a better resolved slat cove

vortex with the Fluent mesh compared to the ANSA mesh.

Figure 21. JSM ANSA-Mesh, Eddy Viscosity Ratio contours near wing root slat cove at α = 4.360.

Figure 22. JSM Fluent-Mesh, Eddy Viscosity Ratio contours near wing root slat cove at α = 4.360.

Figure 23. JSM ANSA-Mesh, Eddy Viscosity Ratio contours near wing tip slat cove at α = 4.360.

20

Figure 24. JSM Fluent-Mesh, Eddy Viscosity Ratio contours near wing tip slat cove at α = 4.360.

Similarly, Figures 23 & 24 show eddy viscosity contours and mesh resolution at sectional cut planes

near wing tip. Again, the slat cove vortex is slightly better resolved by the Fluent mesh, as indicated

by slightly higher predicted eddy viscosity ratio compared to using ANSA mesh. Conversely, the

flow around the wing and in the wing wake regions are better resolved by the ANSA mesh, as

indicated by higher values and less spreading of eddy viscosity contours compared to the Fluent

mesh.

Figure 25. JSM ANSA-Mesh, Eddy Viscosity Ratio contours, wing span, zoom-out α = 4.360.

Figure 26. JSM Fluent-Mesh, Eddy Viscosity Ratio contours, wing span, zoom-out α = 4.360.

21

Figure 27. JSM ANSA-Mesh, Eddy Viscosity Ratio contours, wing span, zoom-in α = 4.360.

Figure 28. JSM Fluent-Mesh, Eddy Viscosity Ratio contours, wing span, zoom-in α = 4.360.

Figure 25 to 28 show several views of the wing wake mesh refinements and eddy viscosity contours

at different sections. Sufficient wake refinement is very important for predictions of lift and drag.

Both ANSA and Fluent meshes have been refined in the wing wake, but each is still found wanting

in different regards. The ANSA mesh has good wake resolution over the wing and few chord lengths

behind the flap trailing edge, but then coarsens quite abruptly. The Fluent mesh, on the other hand,

could be more refined over the wing and immediately downstream, but then has better resolution

further downstream.

Figure 29. JSM ANSA-Mesh, Eddy Viscosity Ratio contours capturing wing wake α = 4.360.

22

Figure 30. JSM Fluent-Mesh, Eddy Viscosity Ratio contours capturing wing wake α = 4.360.

Figures 29 & 30 show the overall wing span wake originating from slat-tracks and passing over

wing and flap using both the ANSA and Fluent meshes and highlight the development and

propagation of wakes on multiple sectional cuts along the chord. Here again, the higher resolution

of the ANSA mesh in the more immediate vicinity of the wing leads to better capturing of the

wakes compared to solutions using the Fluent mesh.

Figure 31. JSM ANSA-Mesh, Eddy Viscosity Ratio contours capturing wing-tip wake α = 4.360.

Figure 32. JSM Fluent-Mesh, Eddy Viscosity Ratio contours capturing wing-tip wake α = 4.360.

Figure 31 & 32 shows the flow development around the wing-tip, with vortices originating from

the slat track and the gap between slat and wing. The ANSA mesh with more mesh refinement in

this area captures these more accurately compared to the Fluent mesh, as the eddy viscosity

contours show.

23

Figure 33. JSM ANSA-Mesh, surface mesh (left) and intermittency contours (right) at α = 4.360.

Figure 34. JSM Fluent-Mesh, surface mesh (left) and intermittency contours (right) at α = 4.360.

The SST-Transition a1=1 intermittency contours predicted with the two meshes at α = 4.360 are

shown in Figures 33 & 34. The laminar-turbulence transition appears to be better resolved using the

Fluent mesh than with the ANSA mesh, owing to the greater mesh refinement in this region. This is

especially noticeable near the leading edge at the mid-span wing location.

Overall, these comparisons of the flow solutions obtained at α = 4.360 with the SST a1=1 and the SST-

Transition a1=1 turbulence model on the two different meshes give a number of insights into the mesh

requirements for capturing the complex flow physics of high lift configurations accurately, some of

which would represent a combination of the resolution used in the two meshes.

24

6 Conclusion and Further Work

The JSM model was simulated with an ANSYS Fluent Meshing mesh using SST-Transition with

a1=1 and SST with a1=1 at six angles of attack (α). Both the computational results with SST a1=1

and SST-Transition a1=1 show good agreement with the experimental data, but both predict a lower

stall angle when compared to experiment. Large differences between wind tunnel and

computational results were observed for drag in both models, especially with increasing angles of

attack. These differences are believed to be primarily due to insufficient grid resolution and overly

rapid decay of turbulent flow structures in key flow regions, such as high-lift device trailing edge

wakes, support bracket wakes, wing-body junction flow, slat and flap cove vortices, flap and wing

tip vortices and any other secondary spiral vortices. The inherent limitation of linear eddy-viscosity

turbulence models which neglect the non-linear effects of turbulent, secondary and swirling flows

may also be a contributing factor. The experimental oil flow and china clay visualizations provided

a more detailed picture of the complex flow patterns around the high-lift system, and highlighted

differences between the computations and experiments. Predictions of attachment line transition,

wing-body junction flow, wing-tip flow and outboard wing separation were seen to differ notably

times from some experimental results. However, computational surface flow patterns on the high-

lift system that included the slat tracks and FTFs, showed qualitatively good agreement with

experimental oil flow visualizations. Correctly predicting viscous wake interactions (especially past

the slat tracks) and wing-tip vortices were key to the overall lift and drag calculations. A clear and

significant difference to experimental results was the computational prediction of a large area of

separation on the outboard wing that already occurred in both cases at α = 14.540, a significantly

lower angle of attack than seen in experiments. The causes for this early prediction of separation

needs to be studied more thoroughly.

In general, further investigation has been identified as necessary for various aspects of modelling

the complex flow physics around such multi-element high lift configurations. For one, results show

that laminar-turbulent transition has a significant and important effect, as also seen in the china

clay visualizations from the experiments, and therefore needs to be correctly accounted for. Also,

the required mesh resolution for grid converged results still needs to be established, taking into

consideration all key areas, which include not only boundary layers and geometric features, but

also flow-interior regions likes wakes, wing tip vortices, cavities, and gaps.

7 Acknowledgement

The authors would like to thank the HiLiftPW3 organizing team for providing rich data for CFD

solver comparisons. Also, the authors would like to thank David Whitaker from CRAY for his

support, and generously recognize CRAY HPC for providing computational resources for

performing these simulations. Many thanks to Florian Menter and Boris Makarov both from

ANSYS, for reviewing the results and providing valuable inputs. Special thanks to Vinod Mahale

and Gandhar Parkhi from ANSYS for assistance during volume mesh generation in ANSYS Fluent

Meshing. Finally, thanks to Aditya Mukane and Udo Tremel, also from ANSYS, for providing

timely help in geometry preparation in ANSYS SpaceClaim Direct Modeler.

25

Fluent Scalability on CRAY XC Series Supercomputers

The Cray XC system offers excellent parallel performance for ANSYS Fluent, with continued

scaling to more than 2,000 cores for ~165-million-cell simulation, as seen in Figure . Cray and ANSYS

are committed to delivering high performance computing capabilities that quickly bring aerospace

applications to new heights of simulation fidelity. This project is just one example of how ANSYS and

Cray collaborate to build robust solutions for a broad set of engineering simulations.

Cray XC40 system combines the advantages of its Aries™ interconnect and Dragonfly network

topology, Intel® Xeon® processors, integrated storage solutions, and major enhancements to the

Cray Linux® Environment and programming environment. The Cray XC40 supercomputer is a

groundbreaking architecture upgradable to 100 petaflops per system.

Figure 35 Performance chart of ANSYS Fluent Simulation on CRAY XC40 Supercomputers.

References

[1] Meredith, P. T., “Viscous Phenomena Affecting High-Lift Systems and Suggestions for Future

CFD Development,” High-Lift Systems Aerodynamics, AGARD CP-515, Sept. 1993, pp. 19-1–

19-8.

[2] Mitsuhiro Murayama, Kazuomi Yamamoto, and Kunihiko Kobayashi. "Validation of

Computations Around High-Lift Configurations by Structured- and Unstructured-

Mesh", Journal of Aircraft, Vol. 43, No. 2 (2006), pp. 395-406.

[3] Murayama, Mitsuhiro Murayama, Yuzuru Yokokawa, Kazuomi Yamamoto, and Yoshine Ueda.

"CFD Validation Study for a High-Lift Configuration of a Civil Aircraft Model", 25th AIAA

Applied Aerodynamics Conference, 25 - 28 June 2007, Miami, FL

[4] 3rd AIAA CFD High Lift Prediction Workshop, URL: http://hiliftpw.larc.nasa.gov/, email:

[email protected], June 2015

[5] Menter, F. R., "Two-Equation Eddy-Viscosity Turbulence Models for Engineering Applications,"

26

AIAA Journal, Vol. 32, No. 8, pp. 1598-1605, (1994).

[6] Menter, F. R., Kuntz, M., and Langtry, R., "Ten Years of Industrial Experience with the SST

Turbulence Model," Turbulence, Heat and Mass Transfer 4, ed: K. Hanjalic, Y. Nagano, and M.

Tummers, Begell House, Inc., 2003, pp. 625 - 632.

[7] Langtry, R. B. and Menter, F. R., "Correlation-Based Transition Modeling for Unstructured

Parallelized Computational Fluid Dynamics Codes," AIAA Journal, Vol. 47, No. 12, December

2009, pp. 2894-2906.

[8] Hiroshi Kato, Keiichi Ishiko, and Akira Yoshizawa. "Optimization of Parameter Values in the

Turbulence Model Aided by Data Assimilation", AIAA Journal, Vol. 54, No. 5 (2016), pp. 1512-

1523.

[9] Ito, T., Ura, H., Yokokawa, Y., Kato, H., Mitsuo, K., and Yamamoto, K., “High-Lift Device

Testing in JAXA 6.5m×5.5m Low-speed Wind Tunnel,” AIAA Paper 2006-3643, June 2006.

[10] 5Ito, T., Ura, H., Yokokawa, Y., Kato, H., Mitsuo, K., and Yamamoto, K., “Experimental and

CFD of a High-Lift Configuration Civil Transport Aircraft Model,” AIAA Paper 2006-3452, June

2006.

[11] C. P. van Dam, “The aerodynamic design of multi-element high-lift systems for Transport

Airplanes,” Progress in Aerospace Science, Vol.38 101-144, 2002.

[12] Rumsey, C., Slotnick, J., Long, M., Stuever, R., Wayman, T., “Summary of the First AIAA CFD

High-Lift Prediction Workshop,” Journal of Aircraft, Vol. 48, No. 6, pp. 2068-2079, November-

December 2011.

[13] Christopher L. Rumsey and Jeffrey P. Slotnick. "Overview and Summary of the Second AIAA

High-Lift Prediction Workshop", Journal of Aircraft, Vol. 52, No. 4 (2015), pp. 1006-1025.

[14] Chaffin, M. S. and Pirzaheh, S., “Unstructured Navier-Stokes High-Lift Computations on a

Trapezoidal Wing,” AIAA Paper 2005-5084, June 2005.

[15] Yokokawa, Y., Murayama, M., Ito, T., and Yamamoto, K., “Experimental and CFD of a High-

Lift Configuration Civil Transport Aircraft Model,” AIAA Paper 2006-3452, June 2006.

[16] Smirnov, P., and Menter, F. R., “Sensitization of the SST turbulence model to rotation and

curvature by applying the Spalart–Shur correction term,” Journal of Turbomachinery, Vol 131,

Issue 4, July 2009.

[17] Lei, Z., “Effect of RANS Turbulence Models on Computational of Separated Flows over a

Wing-Body Configuration,” Transactions of the Japan Society for Aeronautical and Space

Sciences, Vol. 48, Nov. 2005, pp. 150-160.

[18] Rogers, S. E., Roth, K., and Nash, S. M., “CFD Validation of High-Lift Flows with Significant

Wind-Tunnel Effects,” AIAA Paper 2000-4218, Aug. 2000.

[19] Krumbein, A., “Automatic Transition Prediction and Application to 3D High-Lift

Configurations,” AIAA Paper 2006-3164, June 2006.

[20] Perraud, J. and Moens, F., “Transport Aircraft 3D High Lift Wing Numerical Transition

Prediction,” AIAA Paper 2007-264, Jan. 2007.

[21] Krishna, Z., Shoaib, S., John, S., Balasubramanyam, S., and Patrick, S., “ANSYS CFD Study

for High-Lift Aircraft Configurations”. Special Session: High Lift Prediction Using CFD, AIAA

AVIATION, June 2018. Submitted.

ansys cfd validation for civil transport aircraft in high- lift ...cleanup and part-management is...

Documents